Frontiers in Human Neuroscience (Oct 2012)

Parallel ICA identifies sub-components of resting state networks that covary with behavioral indices

  • Timothy eMeier,
  • Joseph eWildenberg,
  • Joseph eWildenberg,
  • Jingyu eLiu,
  • Jingyu eLiu,
  • Jiayu eChen,
  • Jiayu eChen,
  • Vince eCalhoun,
  • Vince eCalhoun,
  • Bharat eBiswal,
  • Mary eMeyerand,
  • Mary eMeyerand,
  • Mary eMeyerand,
  • Rasmus eBirn,
  • Rasmus eBirn,
  • Rasmus eBirn,
  • Vivek ePrabhakaran,
  • Vivek ePrabhakaran,
  • Vivek ePrabhakaran,
  • Vivek ePrabhakaran

DOI
https://doi.org/10.3389/fnhum.2012.00281
Journal volume & issue
Vol. 6

Abstract

Read online

Parallel Independent Component Analysis (para-ICA) is a multivariate method that can identify complex relationships between different data modalities by simultaneously performing Independent Component Analysis on each data set while finding mutual information between the two data sets. We use para-ICA to test the hypothesis that spatial sub-components of common resting state networks (RSNs) covary with specific behavioral measures. Resting state scans and a battery of behavioral indices were collected from 24 younger adults. Group ICA was performed and common RSNs were identified by spatial correlation to publically available templates. Nine RSNs were identified and para-ICA was run on each network with a matrix of behavioral measures serving as the second data type. Five networks had spatial sub-components that significantly correlated with behavioral components. These included a sub-component of the temporo-parietal attention network that differentially covaried with different trial-types of a sustained attention task, sub-components of default mode networks that covaried with attention and working memory tasks, and a sub-component of the bilateral frontal network that split the left inferior frontal gyrus into three clusters according to its cytoarchitecture that differentially covaried with working memory performance. Additionally, we demonstrate the validity of para-ICA in cases with unbalanced dimensions using simulated data.

Keywords